The predictive power of local properties of financial networks

Research paper by Petre Caraiani

Indexed on: 08 Oct '16Published on: 26 Aug '16Published in: Physica A: Statistical Mechanics and its Applications


The literature on analyzing the dynamics of financial networks has focused so far on the predictive power of global measures of networks like entropy or index cohesive force. In this paper, I show that the local network properties have similar predictive power. I focus on key network measures like average path length, average degree or cluster coefficient, and also consider the diameter and the s-metric. Using Granger causality tests, I show that some of these measures have statistically significant prediction power with respect to the dynamics of aggregate stock market. Average path length is most robust relative to the frequency of data used or specification (index or growth rate). Most measures are found to have predictive power only for monthly frequency. Further evidences that support this view are provided through a simple regression model.

Figure 10.1016/j.physa.2016.08.032.0.jpg
Figure 10.1016/j.physa.2016.08.032.1.jpg
Figure 10.1016/j.physa.2016.08.032.2.jpg
Figure 10.1016/j.physa.2016.08.032.3.jpg
Figure 10.1016/j.physa.2016.08.032.4.jpg
Figure 10.1016/j.physa.2016.08.032.5.jpg
Figure 10.1016/j.physa.2016.08.032.6.jpg